Summary:
When there are multiple instances of a forward decl record type, only the first one is emitted with a type index, because
the type is added to a map with a null type index. Avoid this by reordering so that forward decl types aren't added to the map.
Reviewers: rnk
Subscribers: aprantl, hiraditya, arphaman, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D61460
llvm-svn: 360101
The problem was that we were creating a CMOV64rr <TargetFrameIndex>, <TargetFrameIndex>. The entire point of a TFI is that address code is not generated, so there's no way to legalize/lower this. Instead, simply prevent it's creation.
Arguably, we shouldn't be using *Target*FrameIndices in StatepointLowering at all, but that's a much deeper change.
llvm-svn: 360090
The fneg double case is falling back to a subsd in 32-bit mode if you write a test that doesn't trigger a fast-isel abort on the return value.
The subsd lowering has different behavior with respect to nans than using an xor. This is inconsisent with what we would do in SelectionDAG
and can lead to differences between -O0 and -O2.
llvm-svn: 360088
We require d/q suffixes on the memory form of these instructions to disambiguate the memory size.
We don't require it on the register forms, but need to support parsing both with and without it.
Previously we always printed the d/q suffix on the register forms, but it's redundant and
inconsistent with gcc and objdump.
After this patch we should support the d/q for parsing, but not print it when its unneeded.
llvm-svn: 360085
It's possible to use the 'y' mmx constraint with a type narrower than 64-bits.
This patch supports this by bitcasting the mmx type to 64-bits and then
truncating to the desired type.
There are probably other missing type combinations we need to support, but this
is the case we have a bug report for.
Fixes PR41748.
Differential Revision: https://reviews.llvm.org/D61582
llvm-svn: 360069
Reverts "[X86] Remove (V)MOV64toSDrr/m and (V)MOVDI2SSrr/m. Use 128-bit result MOVD/MOVQ and COPY_TO_REGCLASS instead"
Reverts "[TargetLowering][AMDGPU][X86] Improve SimplifyDemandedBits bitcast handling"
Eric Christopher and Jorge Gorbe Moya reported some issues with these patches to me off list.
Removing the CodeGenOnly instructions has changed how fneg is handled during fast-isel with sse/sse2. We're now emitting fsub -0.0, x instead
moving to the integer domain(in a GPR), xoring the sign bit, and then moving back to xmm. This is because the fast isel table no longer
contains an entry for (f32/f64 bitcast (i32/i64)) so the target independent fneg code fails. The use of fsub changes the behavior of nan with
respect to -O2 codegen which will always use a pxor. NOTE: We still have a difference with double with -m32 since the move to GPR doesn't work
there. I'll file a separate PR for that and add test cases.
Since removing the CodeGenOnly instructions was fixing PR41619, I'm reverting r358887 which exposed that PR. Though I wouldn't be surprised
if that bug can still be hit independent of that.
This should hopefully get Google back to green. I'll work with Simon and other X86 folks to figure out how to move forward again.
llvm-svn: 360066
This is a subset of the original commit from rL359879
which was reverted because it could crash when using the 'RemovedInstructions'
structure that enables delayed deletion of dead instructions. The motivating
compile-time win does not require that change though. We should get most of
that win from this change alone.
Using/updating a dominator tree to match math overflow patterns may be very
expensive in compile-time (because of the way CGP uses a DT), so just handle
the single-block case.
See post-commit thread for rL354298 for more details:
http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20190422/646276.html
Differential Revision: https://reviews.llvm.org/D61075
llvm-svn: 359969
Using/updating a dominator tree to match math overflow patterns may be very
expensive in compile-time (because of the way CGP uses a DT), so just handle
the single-block case.
Also, we were restarting the iterator loops when doing the overflow intrinsic
transforms by marking the dominator tree for update. That was done to prevent
iterating over a removed instruction. But we can postpone the deletion using
the existing "RemovedInsts" structure, and that means we don't need to update
the DT.
See post-commit thread for rL354298 for more details:
http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20190422/646276.html
Differential Revision: https://reviews.llvm.org/D61075
llvm-svn: 359879
This is the second part of the commit fixing PR38917 (hoisting
partitially redundant machine instruction). Most of PRE (partitial
redundancy elimination) and CSE work is done on LLVM IR, but some of
redundancy arises during DAG legalization. Machine CSE is not enough
to deal with it. This simple PRE implementation works a little bit
intricately: it passes before CSE, looking for partitial redundancy
and transforming it to fully redundancy, anticipating that the next
CSE step will eliminate this created redundancy. If CSE doesn't
eliminate this, than created instruction will remain dead and eliminated
later by Remove Dead Machine Instructions pass.
The third part of the commit is supposed to refactor MachineCSE,
to make it more clear and to merge MachinePRE with MachineCSE,
so one need no rely on further Remove Dead pass to clear instrs
not eliminated by CSE.
First step: https://reviews.llvm.org/D54839
Fixes llvm.org/PR38917
Reviewers: RKSimon
Subscribers: hfinkel, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D56772
llvm-svn: 359870
The default impementation in the base class for TargetLowering::getRegForInlineAsmConstraint doesn't work for mask registers when the VT is a scalar type integer types since the only legal mask types are vXi1. So we end up just getting whatever the first register class that contains the register. Currently this appears to be VK1, but its really dependent on the order tablegen outputs the register classes.
Some code in the caller ends up looking up the type for this register class and find v1i1 then generates a copyfromreg from the physical k-register with the v1i1 type. Then it generates an any_extend from v1i1 to the scalar VT which isn't legal. This bad any_extend sticks around until isel where it selects a MOVZX32rr8 with a v1i1 input or maybe a i8 input. Not sure but eventually we pick up a copy from VK1 to GR8 in MachineIR which isn't supported. This leads to a failure in physical register copying.
This patch uses the scalar type to find a VK class of the right size. In the attached test case this will be VK16. This causes a bitcast from vk16 to i16 to be generated instead of an any_extend. This will be properly iseled to a VK16 to GR32 copy and a GR32->GR16 extract_subreg.
Fixes PR41678
Differential Revision: https://reviews.llvm.org/D61453
llvm-svn: 359837
The original patch was committed at rL359398 and reverted at rL359695 because of
infinite looping.
This includes a fix to check for a vector splat of "1.0" to avoid the infinite loop.
Original commit message:
This was originally part of D61028, but it's an independent diff.
If we try the repeated divisor reciprocal transform before producing an estimate sequence,
then we have an opportunity to use scalar fdiv. On x86, the trade-off is 1 divss vs. 5
vector FP ops in the default estimate sequence. On recent chips (Skylake, Ryzen), the
full-precision division is only 3 cycle throughput, so that's probably the better perf
default option and avoids problems from x86's inaccurate estimates.
The last 2 tests show that users still have the option to override the defaults by using
the function attributes for reciprocal estimates, but those patterns are potentially made
faster by converting the vector ops (including ymm ops) to scalar math.
Differential Revision: https://reviews.llvm.org/D61149
llvm-svn: 359793
Limiting scalar hadd/hsub generation to the lowest xmm looks to be unnecessary - we will be extracting one upper xmm whatever, and we can remove a shuffle by using the hop which is inline with what shouldUseHorizontalOp expects to happen anyway.
Testing on btver2 (the main target for fast-hops) shows this is beneficial even for float ops where we have a 'shuffle' to extract the float result:
https://godbolt.org/z/0R-U-K
Differential Revision: https://reviews.llvm.org/D61426
llvm-svn: 359786
We already perform horizontal add/sub if we extract from elements 0 and 1, this patch extends it to non-0/1 element extraction indices (as long as they are from the lowest 128-bit vector).
Differential Revision: https://reviews.llvm.org/D61263
llvm-svn: 359707
This is an alternative to D59669 which more aggressively extracts i1 elements from vXi1 bool vectors using a MOVMSK.
Differential Revision: https://reviews.llvm.org/D61189
llvm-svn: 359666
We use both -long-option and --long-option in tests. Switch to --long-option for consistency.
In the "llvm-readelf" mode, -long-option is discouraged as it conflicts with grouped short options and it is not accepted by GNU readelf.
While updating the tests, change llvm-readobj -s to llvm-readobj -S to reduce confusion ("s" is --section-headers in llvm-readobj but --symbols in llvm-readelf).
llvm-svn: 359649
The reordering can leave at least a dead TokenFactor in the graph. This cause the linearize scheduler to fail with something like the assert seen in PR22614. This is only one of many ways we can break the linearize scheduler today so I can't say for sure that any of the other failures in that bug were caused by this issue.
This takes the heavy hammer approach of just running RemoveDeadNodes unconditionally at the end of the PreprocessISelDAG. If this turns out to be a compile time hit, we can try to refine it.
Differential Revision: https://reviews.llvm.org/D61164
llvm-svn: 359582
This removes some of the class variables. Merge basic block processing into
runOnMachineFunction to keep the flags local.
Pass MachineBasicBlock around instead of an iterator. We can get the iterator in
the few places that need it. Allows a range-based outer for loop.
Separate the Atom optimization from the rest of the optimizations. This allows
fixupIncDec to create INC/DEC and still allow Atom to turn it back into LEA
when profitable by its heuristics.
I'd like to improve fixupIncDec to turn LEAs into ADD any time the base or index
register is equal to the destination register. This is profitable regardless of
the various slow flags. But again we would want Atom to be able to undo that.
Differential Revision: https://reviews.llvm.org/D60993
llvm-svn: 359581
We don't have this restriction in IR, so it should not be here
either simply out of consistency. Code that wants to handle FP
exceptions is expected to use the 'strict' variants of these
nodes.
We don't get the frem case because frem by 0.0 produces NaN (invalid),
and that's the remaining check here (so the removed check for frem
was dead code AFAIK).
This is the only place in SDAG that uses "HasFPExceptions", so I
think we should remove that entirely as a follow-up patch.
llvm-svn: 359566
Current LLVM uses pxor+pinsrb on SSE4+ for INSERT_VECTOR_ELT(ZeroVec, 0, Elt) insead of much simpler movd.
INSERT_VECTOR_ELT(ZeroVec, 0, Elt) is idiomatic construct which is used e.g. for _mm_cvtsi32_si128(Elt) and for lowest element initialization in _mm_set_epi32.
So such inefficient lowering leads to significant performance digradations in ceratin cases switching from SSSE3 to SSE4.
https://bugs.llvm.org/show_bug.cgi?id=41512
Here INSERT_VECTOR_ELT(ZeroVec, 0, Elt) is simply converted to SCALAR_TO_VECTOR(Elt) when applicable since latter is closer match to desired behavior and always efficiently lowered to movd and alike.
Committed on behalf of @Serge_Preis (Serge Preis)
Differential Revision: https://reviews.llvm.org/D60852
llvm-svn: 359545
The PrologEpilogInserter need to insert a DW_OP_deref_size before
prepending a memory location expression to an already implicit
expression to avoid having the existing expression act on the memory
address instead of the value behind it.
The reason for using DW_OP_deref_size and not plain DW_OP_deref is that
big-endian targets need to read the right size as simply truncating a
larger read would yield the wrong result (LSB bytes are not at the lower
address).
This re-commit fixes issues reported in the first one. Namely deref was
inserted under wrong conditions and additionally the deref_size argument
was incorrectly encoded.
Differential Revision: https://reviews.llvm.org/D59687
llvm-svn: 359535
This is necessary since SVN r330706, as tail merging can include
CFI instructions since then.
This fixes PR40322 and PR40012.
Differential Revision: https://reviews.llvm.org/D61252
llvm-svn: 359496
Add target shuffle decoding to isHorizontalBinOp as well as ISD::VECTOR_SHUFFLE support.
This does mean we can go through bitcasts so we need to bitcast the extracted args to ensure they are the correct type
Fixes PR39936 and should help with PR39920/PR39921
Differential Revision: https://reviews.llvm.org/D61245
llvm-svn: 359491
Summary:
Extract the logic for doing reassociations
from DAGCombiner::reassociateOps into a helper
function DAGCombiner::reassociateOpsCommutative,
and use that helper to trigger reassociation
on the original operand order, or the commuted
operand order.
Codegen is not identical since the operand order will
be different when doing the reassociations for the
commuted case. That causes some unfortunate churn in
some test cases. Apart from that this should be NFC.
Reviewers: spatel, craig.topper, tstellar
Reviewed By: spatel
Subscribers: dmgreen, dschuff, jvesely, nhaehnle, javed.absar, sbc100, jgravelle-google, hiraditya, aheejin, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D61199
llvm-svn: 359476
Some of the combines might be further improved if we lower more shuffles with X86ISD::VPERMV3 directly, instead of waiting to combine the results.
llvm-svn: 359400
This was originally part of D61028, but it's an independent diff.
If we try the repeated divisor reciprocal transform before producing an estimate sequence,
then we have an opportunity to use scalar fdiv. On x86, the trade-off is 1 divss vs. 5
vector FP ops in the default estimate sequence. On recent chips (Skylake, Ryzen), the
full-precision division is only 3 cycle throughput, so that's probably the better perf
default option and avoids problems from x86's inaccurate estimates.
The last 2 tests show that users still have the option to override the defaults by using
the function attributes for reciprocal estimates, but those patterns are potentially made
faster by converting the vector ops (including ymm ops) to scalar math.
Differential Revision: https://reviews.llvm.org/D61149
llvm-svn: 359398
An xor reduction of a bool vector can be optimized to a parity check of the MOVMSK/BITCAST'd integer - if the population count is odd return 1, else return 0.
Differential Revision: https://reviews.llvm.org/D61230
llvm-svn: 359396
Summary:
The register form of these instructions are CodeGenOnly instructions that cover
GR32->FR32 and GR64->FR64 bitcasts. There is a similar set of instructions for
the opposite bitcast. Due to the patterns using bitcasts these instructions get
marked as "bitcast" machine instructions as well. The peephole pass is able to
look through these as well as other copies to try to avoid register bank copies.
Because FR32/FR64/VR128 are all coalescable to each other we can end up in a
situation where a GR32->FR32->VR128->FR64->GR64 sequence can be reduced to
GR32->GR64 which the copyPhysReg code can't handle.
To prevent this, this patch removes one set of the 'bitcast' instructions. So
now we can only go GR32->VR128->FR32 or GR64->VR128->FR64. The instruction that
converts from GR32/GR64->VR128 has no special significance to the peephole pass
and won't be looked through.
I guess the other option would be to add support to copyPhysReg to just promote
the GR32->GR64 to a GR64->GR64 copy. The upper bits were basically undefined
anyway. But removing the CodeGenOnly instruction in favor of one that won't be
optimized seemed safer.
I deleted the peephole test because it couldn't be made to work with the bitcast
instructions removed.
The load version of the instructions were unnecessary as the pattern that selects
them contains a bitcasted load which should never happen.
Fixes PR41619.
Reviewers: RKSimon, spatel
Reviewed By: RKSimon
Subscribers: hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D61223
llvm-svn: 359392
As predicate masks are legal on AVX512 targets, we avoid MOVMSK in these cases, but we can just bitcast the bool vector to the integer equivalent directly - avoiding expansion of the reduction to a shuffle pattern.
llvm-svn: 359386
AND/OR/XOR tests for the @llvm.experimental.vector.reduce intrinsics
AND/OR are pretty good (pre-AVX512), XOR (not so common but used for parity reduction) is still pretty bad.
llvm-svn: 359385
Fixes PR40332 in the limited case where we're selecting between a target shuffle and a zero vector.
We can extend this in the future to handle more opcodes and non-zero selections.
llvm-svn: 359378
Summary: If we have SSE2 we can use a MOVQ to store 64-bits and avoid falling back to a cmpxchg8b loop. If its a seq_cst store we need to insert an mfence after the store.
Reviewers: spatel, RKSimon, reames, jfb, efriedma
Reviewed By: RKSimon
Subscribers: hiraditya, dexonsmith, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D60546
llvm-svn: 359368
Summary:
Targets like ARM, MSP430, PPC, and SystemZ have complex behavior when
printing the address of a MachineOperand::MO_GlobalAddress. Move that
handling into a new overriden method in each base class. A virtual
method was added to the base class for handling the generic case.
Refactors a few subclasses to support the target independent %a, %c, and
%n.
The patch also contains small cleanups for AVRAsmPrinter and
SystemZAsmPrinter.
It seems that NVPTXTargetLowering is possibly missing some logic to
transform GlobalAddressSDNodes for
TargetLowering::LowerAsmOperandForConstraint to handle with "i" extended
inline assembly asm constraints.
Fixes:
- https://bugs.llvm.org/show_bug.cgi?id=41402
- https://github.com/ClangBuiltLinux/linux/issues/449
Reviewers: echristo, void
Reviewed By: void
Subscribers: void, craig.topper, jholewinski, dschuff, jyknight, dylanmckay, sdardis, nemanjai, javed.absar, sbc100, jgravelle-google, eraman, kristof.beyls, hiraditya, aheejin, kbarton, fedor.sergeev, jrtc27, atanasyan, jsji, llvm-commits, kees, tpimh, nathanchance, peter.smith, srhines
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D60887
llvm-svn: 359337
As detailed on PR40758, Bobcat/Jaguar can perform vector immediate shifts on the same pipes as vector ANDs with the same latency - so it doesn't make sense to replace a shl+lshr with a shift+and pair as it requires an additional mask (with the extra constant pool, loading and register pressure costs).
Differential Revision: https://reviews.llvm.org/D61068
llvm-svn: 359293
A small step towards combining shuffles across vector sizes - this recognizes when a shuffle's operands are all extracted from the same larger source and tries to combine to an unary shuffle of that source instead. Fixes one of the test cases from PR34380.
Differential Revision: https://reviews.llvm.org/D60512
llvm-svn: 359292
We had special case handling here, but it uses a scalar any_extend for the
promotion then bitcasts to the final type. This won't split up the input data
into multiple promoted elements like we need.
This patch falls back to doing the conversion through memory.
Fixes PR41594 which I believe was reflected in the bitcast-vector-bool.ll
changes. The changes to vector-half-conversions.ll are fixing a previously
unknown miscompile from this issue.
Differential Revision: https://reviews.llvm.org/D61114
llvm-svn: 359219
Truncate the movmsk scalar integer result to the equivalent scalar integer width as before but then bitcast to the requested type.
We still have the issue identified in PR41594 but D61114 should handle this.
llvm-svn: 359176
Summary:
This emits labels around heapallocsite calls and S_HEAPALLOCSITE debug
info in codeview. Currently only changes FastISel, so emitting labels still
needs to be implemented in SelectionDAG.
Reviewers: rnk
Subscribers: aprantl, hiraditya, cfe-commits, llvm-commits
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D61083
llvm-svn: 359149
If we have a vector FP division with a splatted divisor, use the existing transform
that converts 'x/y' into 'x * (1.0/y)' to allow more conversions. This can then
potentially be converted into a scalar FP division by existing combines (rL358984)
as seen in the tests here.
That can be a potentially big perf difference if scalar fdiv has better timing
(including avoiding possible frequency throttling for vector ops).
Differential Revision: https://reviews.llvm.org/D61028
llvm-svn: 359147
ReplaceAllUsesWith doesn't remove the node that was replaced. So its left around in the graph messing up use counts on other nodes.
One thing to note, is that this isn't valid if the node being deleted is the root node of an LEA match that gets rejected. In that case the node needs to stay alive because the isel table walking code would still have a reference to it that its going to try to match next. I don't think that's the case here though because the nodes being deleted here should be "and", "srl", and "zero_extend" none of which can be the root node of an LEA match.
Differential Revision: https://reviews.llvm.org/D61048
llvm-svn: 359121
If the types don't match, we can't just remove the shuffle.
There may be some other opportunity for optimization here,
but this should prevent the crashing seen in:
https://bugs.llvm.org/show_bug.cgi?id=41414
llvm-svn: 359095
* Add support for uniquing strings in the remark streamer and emitting the string table in the remarks section.
* Add parsing support for the string table in the RemarkParser.
From this remark:
```
--- !Missed
Pass: inline
Name: NoDefinition
DebugLoc: { File: 'test-suite/SingleSource/UnitTests/2002-04-17-PrintfChar.c',
Line: 7, Column: 3 }
Function: printArgsNoRet
Args:
- Callee: printf
- String: ' will not be inlined into '
- Caller: printArgsNoRet
DebugLoc: { File: 'test-suite/SingleSource/UnitTests/2002-04-17-PrintfChar.c',
Line: 6, Column: 0 }
- String: ' because its definition is unavailable'
...
```
to:
```
--- !Missed
Pass: 0
Name: 1
DebugLoc: { File: 3, Line: 7, Column: 3 }
Function: 2
Args:
- Callee: 4
- String: 5
- Caller: 2
DebugLoc: { File: 3, Line: 6, Column: 0 }
- String: 6
...
```
And the string table in the .remarks/__remarks section containing:
```
inline\0NoDefinition\0printArgsNoRet\0
test-suite/SingleSource/UnitTests/2002-04-17-PrintfChar.c\0printf\0
will not be inlined into \0 because its definition is unavailable\0
```
This is mostly supposed to be used for testing purposes, but it gives us
a 2x reduction in the remark size, and is an incremental change for the
updates to the remarks file format.
Differential Revision: https://reviews.llvm.org/D60227
llvm-svn: 359050
The simple case of:
```
int *callee();
void *caller(void *a) {
if (a == NULL)
return callee();
return a;
}
```
would generate a regular call instead of a tail call because we don't
look through the bitcast of the call to `callee` when duplicating the
return blocks.
Differential Revision: https://reviews.llvm.org/D60837
llvm-svn: 359041
In preparation of https://reviews.llvm.org/D60837, add this test where
we don't perform a tail call because we don't look through a bitcast.
llvm-svn: 359040
Circling back to a leftover bit from PR39859:
https://bugs.llvm.org/show_bug.cgi?id=39859#c1
...we have this counter-intuitive (based on the test diffs) opportunity to use 'psubus'.
This appears to be the better perf option for both Haswell and Jaguar based on llvm-mca.
We already do this transform for the SETULT predicate, so this makes the code more
symmetrical too. If we have pminub/pminuw, we prefer those, so this should not affect
anything but pre-SSE4.1 subtargets.
$ cat before.s
movdqa -16(%rip), %xmm2 ## xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
pxor %xmm0, %xmm2
pcmpgtw -32(%rip), %xmm2 ## xmm2 = [255,255,255,255,255,255,255,255]
pand %xmm2, %xmm0
pandn %xmm1, %xmm2
por %xmm2, %xmm0
$ cat after.s
movdqa -16(%rip), %xmm2 ## xmm2 = [256,256,256,256,256,256,256,256]
psubusw %xmm0, %xmm2
pxor %xmm3, %xmm3
pcmpeqw %xmm2, %xmm3
pand %xmm3, %xmm0
pandn %xmm1, %xmm3
por %xmm3, %xmm0
$ llvm-mca before.s -mcpu=haswell
Iterations: 100
Instructions: 600
Total Cycles: 909
Total uOps: 700
Dispatch Width: 4
uOps Per Cycle: 0.77
IPC: 0.66
Block RThroughput: 1.8
$ llvm-mca after.s -mcpu=haswell
Iterations: 100
Instructions: 700
Total Cycles: 409
Total uOps: 700
Dispatch Width: 4
uOps Per Cycle: 1.71
IPC: 1.71
Block RThroughput: 1.8
Differential Revision: https://reviews.llvm.org/D60838
llvm-svn: 358999
If we only match build vectors, we can miss some patterns
that use shuffles as seen in the affected tests.
Note that the underlying calls within getSplatSourceVector()
have the potential for compile-time explosion because of
exponential recursion looking through binop opcodes, but
currently the list of supported opcodes is very limited.
Both of those problems should be addressed in follow-up
patches.
llvm-svn: 358984
Summary:
The DAGCombiner is rewriting (canonicalizing) an ISD::ADD
with no common bits set in the operands as an ISD::OR node.
This could sometimes result in "missing out" on some
combines that normally are performed for ADD. To be more
specific this could happen if we already have rewritten an
ADD into OR, and later (after legalizations or combines)
we expose patterns that could have been optimized if we
had seen the OR as an ADD (e.g. reassociations based on ADD).
To make the DAG combiner less sensitive to if ADD or OR is
used for these "no common bits set" ADD/OR operations we
now apply most of the ADD combines also to an OR operation,
when value tracking indicates that the operands have no
common bits set.
Reviewers: spatel, RKSimon, craig.topper, kparzysz
Reviewed By: spatel
Subscribers: arsenm, rampitec, lebedev.ri, jvesely, nhaehnle, hiraditya, javed.absar, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D59758
llvm-svn: 358965
This was supposed to be NFC, but the change in SDLoc
definitions causes instruction scheduling changes.
There's nothing x86-specific in this code, and it can
likely be used from DAGCombiner's simplifyVBinOp().
llvm-svn: 358930
This patch adds support for BigBitWidth -> SmallBitWidth bitcasts, splitting the DemandedBits/Elts accordingly.
The AMDGPU backend needed an extra (srl (and x, c1 << c2), c2) -> (and (srl(x, c2), c1) combine to encourage BFE creation, I investigated putting this in DAGCombine but it caused a lot of noise on other targets - some improvements, some regressions.
The X86 changes are all definite wins.
Differential Revision: https://reviews.llvm.org/D60462
llvm-svn: 358887
Summary:
If you pass two 1024 bit vectors in IR with AVX2 on Windows 64. Both vectors will be split in four 256 bit pieces. The four pieces of the first argument will be passed indirectly using 4 gprs. The second argument will get passed via pointers in memory.
The PartOffsets stored for the second argument are all in terms of its original 1024 bit size. So the PartOffsets for each piece are 32 bytes apart. So if we consider it for copy elision we'll only load an 8 byte pointer, but we'll move the address 32 bytes. The stack object size we create for the first part is probably wrong too.
This issue was encountered by ISPC. I'm working on getting a reduce test case, but wanted to go ahead and get feedback on the fix.
Reviewers: rnk
Reviewed By: rnk
Subscribers: dbabokin, llvm-commits, hiraditya
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D60801
llvm-svn: 358817
Fix for https://bugs.llvm.org/show_bug.cgi?id=41477. On the x32 ABI
with stack probing a dynamic alloca will result in a WIN_ALLOCA_32
with a 32-bit size. The current implementation tries to copy it into
RAX, resulting in a physreg copy error. Fix this by copying to EAX
instead. Also fix incorrect opcodes or registers used in subs.
llvm-svn: 358807
The MOVZX doesn't require an immediate to be encoded at all. Though it does use
a 2 byte opcode so its the same size as a 1 byte immediate. But it has a
separate source and dest register so can help avoid copies.
llvm-svn: 358805
There's one slight regression in here because we don't check that the immediate
already allowed movzx before the shift. I'll fix that next.
llvm-svn: 358804
Summary:
This emits labels around heapallocsite calls and S_HEAPALLOCSITE debug
info in codeview. Currently only changes FastISel, so emitting labels still
needs to be implemented in SelectionDAG.
Reviewers: hans, rnk
Subscribers: aprantl, hiraditya, cfe-commits, llvm-commits
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D60800
llvm-svn: 358783
combineVectorTruncationWithPACKUS is currently splitting the upper bit bit masking into 128-bit subregs and then concatenating them back together.
This was originally done to avoid regressions that caused existing subregs to be concatenated to the larger type just for the AND masking before being extracted again. This was fixed by @spatel (notably rL303997 and rL347356).
This also lets SimplifyDemandedBits do some further improvements before it hits the recursive depth limit.
My only annoyance with this is that we were broadcasting some xmm masks but we seem to have lost them by moving to ymm - but that's a known issue as the logic in lowerBuildVectorAsBroadcast isn't great.
Differential Revision: https://reviews.llvm.org/D60375#inline-539623
llvm-svn: 358692
This replaces the MOVMSK combine introduced at D52121/rL342326
(movmsk (setne (and X, (1 << C)), 0)) -> (movmsk (X << C))
with the more general icmp lowering so it can pick up more cases through bitcasts - notably vXi8 cases which use vXi16 shifts+masks, this patch can remove the mask and use pcmpgtb(0,x) for the sra.
Differential Revision: https://reviews.llvm.org/D60625
llvm-svn: 358651
The test file has pairs of tests that are logically equivalent:
https://rise4fun.com/Alive/2zQ
%t4 = and i8 %t1, 8
%t5 = zext i8 %t4 to i16
%sh = shl i16 %t5, 2
%t6 = add i16 %sh, %t0
=>
%t4 = and i8 %t1, 8
%sh2 = shl i8 %t4, 2
%z5 = zext i8 %sh2 to i16
%t6 = add i16 %z5, %t0
...so if we can fold the shift op into LEA in the 1st pattern, then we
should be able to do the same in the 2nd pattern (unnecessary 'movzbl'
is a separate bug I think).
We don't want to do this any sooner though because that would conflict
with generic transforms that try to narrow the width of the shift.
Differential Revision: https://reviews.llvm.org/D60789
llvm-svn: 358622
Improves codegen demonstrated by D60512 - instructions represented by X86ISD::PERMV/PERMV3 can never memory fold the operand used for their index register.
This patch updates the 'isUseOfShuffle' helper into the more capable 'isFoldableUseOfShuffle' that recognises that the op is used for a X86ISD::PERMV/PERMV3 index mask and can't be folded - allowing us to use broadcast/subvector-broadcast ops to reduce the size of the mask constant pool data.
Differential Revision: https://reviews.llvm.org/D60562
llvm-svn: 358516
The pattern we replaced these with may be too hard to match as demonstrated by
PR41496 and PR41316.
This patch restores the intrinsics and then we can start focusing
on the optimizing the intrinsics.
I've mostly reverted the original patch that removed them. Though I modified
the avx512 intrinsics to not have masking built in.
Differential Revision: https://reviews.llvm.org/D60674
llvm-svn: 358427
Summary:
Use KnownBits::computeForAddSub/computeForAddCarry
in SelectionDAG::computeKnownBits when doing value
tracking for addition/subtraction.
This should improve the precision of the known bits,
as we only used to make a simple estimate of known
zeroes. The KnownBits support functions are also
able to deduce bits that are known to be one in the
result.
Reviewers: spatel, RKSimon, nikic, lebedev.ri
Reviewed By: nikic
Subscribers: nikic, javed.absar, lebedev.ri, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D60460
llvm-svn: 358372
Apparently there are some stray IMPLICIT_DEF operations that weren't in the
checks. Not sure if they've always been there or something changed at some
point.
llvm-svn: 358371
Other opcodes shouldn't be CSE'd until we can be sure debug info quality won't
be degraded.
This change also improves the IRTranslator so that in most places, but not all,
it creates constants using the MIRBuilder directly instead of first creating a
new destination vreg and then creating a constant. By doing this, the
buildConstant() method can just return the vreg of an existing G_CONSTANT
instead of having to create a COPY from it.
I measured a 0.2% improvement in compile time and a 0.9% improvement in code
size at -O0 ARM64.
Compile time:
Program base cse diff
test-suite...ark/tramp3d-v4/tramp3d-v4.test 9.04 9.12 0.8%
test-suite...Mark/mafft/pairlocalalign.test 2.68 2.66 -0.7%
test-suite...-typeset/consumer-typeset.test 5.53 5.51 -0.4%
test-suite :: CTMark/lencod/lencod.test 5.30 5.28 -0.3%
test-suite :: CTMark/Bullet/bullet.test 25.82 25.76 -0.2%
test-suite...:: CTMark/ClamAV/clamscan.test 6.92 6.90 -0.2%
test-suite...TMark/7zip/7zip-benchmark.test 34.24 34.17 -0.2%
test-suite :: CTMark/SPASS/SPASS.test 6.25 6.24 -0.1%
test-suite...:: CTMark/sqlite3/sqlite3.test 1.66 1.66 -0.1%
test-suite :: CTMark/kimwitu++/kc.test 13.61 13.60 -0.0%
Geomean difference -0.2%
Code size:
Program base cse diff
test-suite...-typeset/consumer-typeset.test 1315632 1266480 -3.7%
test-suite...:: CTMark/ClamAV/clamscan.test 1313892 1297508 -1.2%
test-suite :: CTMark/lencod/lencod.test 1439504 1423112 -1.1%
test-suite...TMark/7zip/7zip-benchmark.test 2936980 2904172 -1.1%
test-suite :: CTMark/Bullet/bullet.test 3478276 3445460 -0.9%
test-suite...ark/tramp3d-v4/tramp3d-v4.test 8082868 8033492 -0.6%
test-suite :: CTMark/kimwitu++/kc.test 3870380 3853972 -0.4%
test-suite :: CTMark/SPASS/SPASS.test 1434904 1434896 -0.0%
test-suite...Mark/mafft/pairlocalalign.test 764528 764528 0.0%
test-suite...:: CTMark/sqlite3/sqlite3.test 782092 782092 0.0%
Geomean difference -0.9%
Differential Revision: https://reviews.llvm.org/D60580
llvm-svn: 358369
We had many tablegen patterns for these instructions. And due to the
commutability of the patterns, tablegen expands them to even more patterns. All
together VPTESTMD patterns accounted for more the 50K of the 610K isel table.
This had gotten bad when we stopped canonicalizing AND to vXi64. This required
a pattern for every combination of bitcast input type.
This change moves the matching to custom code where it is easier to look through
the bitcasts without being concerned with the specific types.
The test changes are because we are now stricter with one use checks as its
required to make load folding legal. We now require the AND and any BITCAST to
only have a single use. This prevents forming VPTESTM and a VPAND with the same
inputs.
We now support broadcast loads for 128/256 patterns without VLX. We'll widen to
512-bit like and still fold the broadcast since the amount of memory read
doesn't change.
There are a few tests that got slightly longer because are now prefering
load + VPTESTM over XOR+VPCMPEQ for (seteq (load), allzeros). Previously we were
able to share the XOR with multiple VPTESTM instructions.
llvm-svn: 358359
We're better of emitting a single compare + kand rather than a compare for the
other use and a masked compare.
I'm looking into using custom instruction selection for VPTESTM to reduce the
ridiculous number of permutations of patterns in the isel table. Putting a one
use check on all masked compare folding makes load fold matching in the custom
code easier.
llvm-svn: 358358
Summary:
The Linux kernel uses PC-relative mode, so allow that when the code model is
"kernel".
Reviewers: craig.topper
Reviewed By: craig.topper
Subscribers: llvm-commits, kees, nickdesaulniers
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D60643
llvm-svn: 358343
// shuffle (concat X, undef), (concat Y, undef), Mask -->
// concat (shuffle X, Y, Mask0), (shuffle X, Y, Mask1)
The ARM changes with 'vtrn' and narrowed 'vuzp' are improvements.
The x86 changes look neutral or better. There's one test with an
extra instruction, but that could be reversed for a subtarget with
the right attributes. But by default, we want to avoid the 256-bit
op when possible (in my motivating benchmark, a handful of ymm ops
sprinkled into a sequence of xmm ops are triggering frequency
throttling on Haswell resulting in significantly worse perf).
Differential Revision: https://reviews.llvm.org/D60545
llvm-svn: 358291
Currently combineHorizontalPredicateResult only handles anyof/allof reduction patterns of legal types, which can be tricky to match as type legalization of bools can introduce bitcasts/truncs/extensions.
This patch extends combineHorizontalPredicateResult to recognise vXi1 bool reductions as well and uses the existing combineBitcastvxi1 helper to create the MOVMSK necessary to then compare the signmask result.
This ensures the accuracy of the reduction costs added in D60403 which assume the MOVMSK generation.
Differential Revision: https://reviews.llvm.org/D60610
llvm-svn: 358286
It causes clang to crash while building Chromium. See https://crbug.com/952230
for reproducer.
> The PrologEpilogInserter need to insert a DW_OP_deref_size before
> prepending a memory location expression to an already implicit
> expression to avoid having the existing expression act on the memory
> address instead of the value behind it.
>
> The reason for using DW_OP_deref_size and not plain DW_OP_deref is that
> big-endian targets need to read the right size as simply truncating a
> larger read would yield the wrong result (LSB bytes are not at the lower
> address).
>
> Differential Revision: https://reviews.llvm.org/D59687
llvm-svn: 358281
The PrologEpilogInserter need to insert a DW_OP_deref_size before
prepending a memory location expression to an already implicit
expression to avoid having the existing expression act on the memory
address instead of the value behind it.
The reason for using DW_OP_deref_size and not plain DW_OP_deref is that
big-endian targets need to read the right size as simply truncating a
larger read would yield the wrong result (LSB bytes are not at the lower
address).
Differential Revision: https://reviews.llvm.org/D59687
llvm-svn: 358268
If the upper bits of the SHL result aren't used, we might be able to use a narrower shift. For example, on X86 this can turn a 64-bit into 32-bit enabling a smaller encoding.
Differential Revision: https://reviews.llvm.org/D60358
llvm-svn: 358257
If the vector setcc has been legalized then we will need to convert a vector boolean of 0 or -1 to a scalar boolean of 0 or 1.
The added test case previously crashed in 32-bit mode by creating a setcc with an i64 condition that type legalization couldn't expand.
llvm-svn: 358218
This patch adds patterns for turning bitcasted atomic load/store into movss/sd.
It also removes the pseudo instructions for atomic RMW fadd. Instead just adding isel patterns for folding an atomic load into addss/sd. And relying on the new movss/sd store pattern to handle the write part.
This also makes the fadd patterns use VEX and EVEX instructions when AVX or AVX512F are enabled.
Differential Revision: https://reviews.llvm.org/D60394
llvm-svn: 358215
With correct test checks this time.
If we have X87, but not SSE2 we can atomicaly load an i64 value into the significand of an 80-bit extended precision x87 register using fild. We can then use a fist instruction to convert it back to an i64 integ
This matches what gcc and icc do for this case and removes an existing FIXME.
llvm-svn: 358214
If we have X87, but not SSE2 we can atomicaly load an i64 value into the significand of an 80-bit extended precision x87 register using fild. We can then use a fist instruction to convert it back to an i64 integer and store it to a stack temporary. From there we can do two 32-bit loads to get the value into integer registers without worrying about atomicness.
This matches what gcc and icc do for this case and removes an existing FIXME.
Differential Revision: https://reviews.llvm.org/D60156
llvm-svn: 358211
// bo (build_vec ...undef, x, undef...), (build_vec ...undef, y, undef...) -->
// build_vec ...undef, (bo x, y), undef...
The lifetime of the nodes in these examples is different for variables versus constants,
but they are all build vectors briefly, so I'm proposing to catch them in this form to
handle all of the leading examples in the motivating test file.
Before we have build vectors, we might have insert_vector_element. After that, we might
have scalar_to_vector and constant pool loads.
It's going to take more work to ensure that FP vector operands are getting simplified
with undef elements, so this transform can apply more widely. In a non-loose FP environment,
we are likely simplifying FP elements to NaN values rather than undefs.
We also need to allow more opcodes down this path. Eg, we don't handle FP min/max flavors
yet.
Differential Revision: https://reviews.llvm.org/D60514
llvm-svn: 358172
foldMaskedShiftToScaledMask tries to reorder and & shl to enable the shl to fold into an LEA. But if there is an any_extend between them it doesn't work.
This patch modifies the code to look through any_extend from i32 to i64 when the and mask only uses bits that weren't from the extended part.
This will prevent a regression from D60358 caused by 64-bit SHL being narrowed to 32-bits when their upper bits aren't demanded.
Differential Revision: https://reviews.llvm.org/D60532
llvm-svn: 358139
If we have an (add X, (and (aext (shl Y, C1)), C2)), we can pull the shift through and+aext to fold into an LEA with the.
Assuming C1 is small enough and C2 masks off all of the extend bits.
This pattern showed up in D60358. And we need to handle it to prevent a regression.
llvm-svn: 358124
Certain optimisations from ConstantHoisting and CGP rely on Selection DAG not
seeing through to the constant in other blocks. Revert this patch while we come
up with a better way to handle that.
I will try to follow this up with some better tests.
llvm-svn: 358113
This lines up with what we do for regular subtract and it matches up better with X86 assumptions in isel patterns that add with immediate is more canonical than sub with immediate.
Differential Revision: https://reviews.llvm.org/D60020
llvm-svn: 358027
When bitcasting from a source op to a larger bitwidth op, split the demanded bits and OR them on top of one another and demand those merged bits in the SimplifyDemandedBits call on the source op.
llvm-svn: 357992
I was looking at a potential DAGCombiner fix for 1 of the regressions in D60278, and it caused severe regression test pain because x86 TLI lies about the desirability of 8-bit shift ops.
We've hinted at making all 8-bit ops undesirable for the reason in the code comment:
// TODO: Almost no 8-bit ops are desirable because they have no actual
// size/speed advantages vs. 32-bit ops, but they do have a major
// potential disadvantage by causing partial register stalls.
...but that leads to massive diffs and exposes all kinds of optimization holes itself.
Differential Revision: https://reviews.llvm.org/D60286
llvm-svn: 357912
Summary:
Previously we would use MOVZXrm8/MOVZXrm16, but those are longer encodings.
This is similar to what we do in the loadi32 predicate.
Reviewers: RKSimon, spatel
Reviewed By: RKSimon
Subscribers: hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D60341
llvm-svn: 357875
In the case where we only want the sign bit (e.g. when using PACKSS truncation of comparison results for MOVMSK) then we can just demand the sign bit of the source operands.
This makes use of the fact that PACKSS saturates out of range values to the min/max int values - so the sign bit is always preserved.
Differential Revision: https://reviews.llvm.org/D60333
llvm-svn: 357859
This function reorders AND and SHL to enable the SHL to fold into an LEA. The
upper bits of the AND will be shifted out by the SHL so it doesn't matter what
mask value we use for these bits. By using sign bits from the original mask in
these upper bits we might enable a shorter immediate encoding to be used.
llvm-svn: 357846
When we shift the AND mask over we should shift in sign bits instead of zero bits. The scale in the LEA will shift these bits out so it doesn't matter whether we mask the bits off or not. Using sign bits will potentially allow a sign extended immediate to be used.
Also add some other test cases for cases that are currently optimal.
llvm-svn: 357845
The expansion of TCRETURNri(64) would not keep operand flags like
undef/renamable/etc. which can result in machine verifier issues.
Also add plumbing to be able to use `-run-pass=x86-pseudo`.
llvm-svn: 357808
Summary:
This avoids needing an isel pattern for each condition code. And it removes translation switches for converting between Jcc instructions and condition codes.
Now the printer, encoder and disassembler take care of converting the immediate. We use InstAliases to handle the assembly matching. But we print using the asm string in the instruction definition. The instruction itself is marked IsCodeGenOnly=1 to hide it from the assembly parser.
Reviewers: spatel, lebedev.ri, courbet, gchatelet, RKSimon
Reviewed By: RKSimon
Subscribers: MatzeB, qcolombet, eraman, hiraditya, arphaman, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D60228
llvm-svn: 357802
Summary:
This avoids needing an isel pattern for each condition code. And it removes translation switches for converting between SETcc instructions and condition codes.
Now the printer, encoder and disassembler take care of converting the immediate. We use InstAliases to handle the assembly matching. But we print using the asm string in the instruction definition. The instruction itself is marked IsCodeGenOnly=1 to hide it from the assembly parser.
Reviewers: andreadb, courbet, RKSimon, spatel, lebedev.ri
Reviewed By: andreadb
Subscribers: hiraditya, lebedev.ri, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D60138
llvm-svn: 357801
Summary:
Reorder the condition code enum to match their encodings. Move it to MC layer so it can be used by the scheduler models.
This avoids needing an isel pattern for each condition code. And it removes
translation switches for converting between CMOV instructions and condition
codes.
Now the printer, encoder and disassembler take care of converting the immediate.
We use InstAliases to handle the assembly matching. But we print using the
asm string in the instruction definition. The instruction itself is marked
IsCodeGenOnly=1 to hide it from the assembly parser.
This does complicate the scheduler models a little since we can't assign the
A and BE instructions to a separate class now.
I plan to make similar changes for SETcc and Jcc.
Reviewers: RKSimon, spatel, lebedev.ri, andreadb, courbet
Reviewed By: RKSimon
Subscribers: gchatelet, hiraditya, kristina, lebedev.ri, jdoerfert, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D60041
llvm-svn: 357800
Second half of PR40800, this patch adds DAG undef handling to fcmp instructions to match the behavior in llvm::ConstantFoldCompareInstruction, this permits constant folding of vector comparisons where some elements had been reduced to UNDEF (by SimplifyDemandedVectorElts etc.).
This involves a lot of tweaking to reduced tests as bugpoint loves to reduce fcmp arguments to undef........
Differential Revision: https://reviews.llvm.org/D60006
llvm-svn: 357765
There are a variety of vector patterns that may be profitably reduced to a
scalar op when scalar ops are performed using a subset (typically, the
first lane) of the vector register file.
For x86, this is true for float/double ops and element 0 because
insert/extract is just a sub-register rename.
Other targets should likely enable the hook in a similar way.
Differential Revision: https://reviews.llvm.org/D60150
llvm-svn: 357760
Summary:
Teach SelectionDAG how to compute known bits of ISD::CopyFromReg if
the virtual reg used has one def only.
This can be particularly useful when calling isBaseWithConstantOffset()
with the ISD::CopyFromReg argument, as more optimizations may get enabled
in the result.
Also add a missing truncation on X86, found by testing of this patch.
Change-Id: Id1c9fceec862d118c54a5b53adf72ada5d6daefa
Reviewers: bogner, craig.topper, RKSimon
Reviewed By: RKSimon
Subscribers: lebedev.ri, nemanjai, jvesely, nhaehnle, javed.absar, jsji, jdoerfert, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D59535
llvm-svn: 357745
We already promote SRL and SHL to i32.
This will introduce sign extends sometimes which might be harder to deal with than the zero we use for promoting SRL. I ran this through some of our internal benchmark lists and didn't see any major regressions.
I think there might be some DAG combine improvement opportunities in the test changes here.
Differential Revision: https://reviews.llvm.org/D60278
llvm-svn: 357743
Lowering safepoint checks that all gc.relocaes observed in safepoint
must be lowered. However Fast-Isel is able to skip dead gc.relocate.
To resolve this issue we just ignore dead gc.relocate in the check.
Reviewers: reames
Reviewed By: reames
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D60184
llvm-svn: 357742
It unnecessarily breaks previously-working code which used varargs,
but didn't pass any float/double arguments (such as EDK2).
Also revert the fixup on top of that:
Revert [X86] Fix a test from r357317
This reverts r357317 (git commit d413f41de6)
This reverts r357380 (git commit 7af32444b9)
llvm-svn: 357718
SUBREG_TO_REG is supposed to be used to assert that we know the upper bits are
zero. But that isn't the case here. We've done no analysis of the inputs.
llvm-svn: 357673
The Fast ISel has a fallback to SelectionDAGISel in case it cannot handle the instruction.
This works as follows:
Using reverse order, try to select instruction using Fast ISel, if it cannot handle instruction it fallbacks to SelectionDAGISel
for these instructions if it is a call and continue fast instruction selections.
However if unhandled instruction is not a call or statepoint related instruction it fallbacks to SelectionDAGISel for all remaining
instructions in basic block.
However gc.result instruction is missed and as a result it is possible that gc.result is processed earlier than statepoint
causing breakage invariant the gc.results should be handled after statepoint.
Test is updated because in the current form fast-isel cannot handle ret instruction (due to i1 ret type without explicit ext)
and as a result test does not check fast-isel at all.
Reviewers: reames
Reviewed By: reames
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D60182
llvm-svn: 357672
These inserters inserted some instructions to zero some registers and copied from virtual registers to physical registers.
This change instead inserts the zeros directly into the DAG at lowering time using new ISD opcodes
that take the extra zeroes as inputs. The zeros will then go through isel on their own to select
the MOV32r0 pseudo. Then we just need to mention the physical registers directly
in the isel patterns and the isel table and InstrEmitter will take care of inserting the necessary
copies to/from physical registers.
llvm-svn: 357659
This custom inserter existed so we could do a weird thing where we pretended that the instructions support
a full address mode instead of taking a pointer in EAX/RAX. I think was largely so we could be pointer
size agnostic in the isel pattern.
To make this work we would then put the address into an LEA into EAX/RAX in front of the instruction after
isel. But the LEA is overkill when we just have a base pointer. So we end up using the LEA as a slower MOV
instruction.
With this change we now just do custom selection during isel instead and just assign the incoming address
of the intrinsic into EAX/RAX based on its size. After the intrinsic is selected, we can let isel take
care of selecting an LEA or other operation to do any address computation needed in this basic block.
I've also split the instruction into a 32-bit mode version and a 64-bit mode version so the implicit
use is properly sized based on the pointer. Without this we get comments in the assembly output about
killing eax and defing rax or vice versa depending on whether we define the instruction to use EAX/RAX.
llvm-svn: 357652
This pattern would show up as a regression if we more
aggressively convert vector FP ops to scalar ops.
There's still a missed optimization for the v4f64 legal
case (AVX) because we create that h-op with an undef operand.
We should probably just duplicate the operands for that
pattern to avoid trouble.
llvm-svn: 357642
Amy Huang